Posts Tagged ‘bigdata’
Posted by Jakub Holý on July 31, 2014
- Video: The Unreasonable Effectiveness of Dynamic Typing for Practical Programs – a static-typing zealot turned friend of dynamic typing under the experience of real-world projects and problems shares thoughts about the limits of type systems (f.ex. both energy and torque are measured in N*m yet cannot be combined) and their cost: according to the Hanenberg’s experiment about static and dynamic typing => the time required to handle the time chacker > time to debug the errors that it would have caught. According to a review of issues at GitHub, only 2% of reported issues for JS, Clojure, Python, and Ruby are type errors and for a large, closed-source Python project type/name/attribute errors were 1%. “I have come to believe that tests are a much better investment [than static typing].” Rigorous type system/model => limited applicability (due to different needs) <=> modelling some things with types doesn’t cut it. “Are the costs of static typing offset by a few percent fewer defects? Is agility more important than reliability?” “Static types are anti-modular” – too a tight coupling. “Static type checking comes at the expense of complexity, brittleness and a tendency to monoliths.”
(Personally I miss static typing – but that is perhaps due to having relied on it for so long.)
- ThoughtWorks Tech Radar July 2014 (pdf): f.ex. Ansible in Adapt, Masterless Chef/Puppet in Trial, Machine image as a build artifact: Trial, PostgreSQL for NoSQL: Trial, Adopt Dropwizard (Rest 4 Java), Go lang, Reactive Extensions across langs [JH: RxJava, RxJS, ..]; Asses Property-based (generative) testing, … . Other highlights: Mapbox (open-source mapping platform), OpenID Connect as a less complex and thus promising alternative to SAML/generic OAuth, Pacto/Pact for Consumer-Driven Contracts (contract => simulate consumers/stubb producers => test your REST clients against the contract so that the rest of tests can assume it is correct and use a stubbed client), Swagger for REST documentation.
- The madness of layered architecture – a nice critique of over-designed “enterprise” apps, why that is a problem (SRP, cost of code, unclear where to do a change, ….), why it is different from the successful layered network stack of Ethernet/IP/TCP/… (because in an app, all layers are on the same level of abstraction); bottom line: do not add a layer unless you have a really good reason (hint: the advice of a consultant/speaker does not count as one)
- Key Takeaway Points and Lessons Learned from QCon New York 2014 (viz @RiczWest) – “[..] deep insights into real-world architectures and state of the art software development practices, from a practioner’s perspective.” – architectures of Fb, Foursquare etc., continuous delivery, creating culture, real world functional programming, … .
- Questioning the Lambda Architecture (J. Kreps of LinkedIn) – maintaining the same processing in two very different systems (one batch, one stream & real-time) is a maintenance nightmare => improve the RT/stream processing to handle re-processing and thus both (using e.g. Kafka to store the data and thus be able to re-play them)
- Google: Checklist for mobile website improvement
- Google Dataflow and the transition from batch to stream processing – G. Dataflow might not be a Hadoop killer due to requiring that the data are in the Google Cloud but the trend is clear, going away from batch processing to more stream-oriented processing with tools like Spark, Flume etc. that are faster thanks to using memory better and more flexible thanks to not being limited to the rigitd two-stage model of map-reduce. (Reportedly, Google – the one that made Map-Reduce popular – doesn’t use it anymore.)
- OS X: Extract JDK to folder, without running installer
Society, economics, people
- HBR: The Power of Meeting Your Employees’ Needs – people feel better, perform better, are more engaged and likely to stay longer (=> profitability) when 4 basic needs are met: physical [energy] renewal (=> give opportunity, encourage to take a nap or do whatever that helps), value – feeling of being valued by the company, ability to focus, purpose (i.e. serving something larger than ourselves). “What’s surprising about our survey’s results is how dramatically and positively getting these needs met is correlated with every variable that influences performance. It would be statistically significant if meeting a given need correlated with a rise of even one or two percentage points in a performance variable such as engagement, or retention. Instead, we found that meeting even one of the four core needs had a dramatic impact on every performance variable we studied. [..] when all four needs are met, the effect on engagement rises from 50% for one need, to 125%. Engagement, in turn, has been positively correlated with profitability. [..] employers with the most engaged employees were 22% more profitable than those with the least engaged employees.”
“[..] those who were encouraged to take intermittent breaks reported they were 50% more engaged, more than twice as likely to stay with the company, and twice as healthy overall. Valuing and encouraging renewal requires no financial investment. What it does require is a willingness among leaders to test their longstanding assumption that that performance is best measured by the number of hours employees puts in – and the more continuous the better — rather than by the value they generate, however they choose to do their work.“
- The Pitchforks Are Coming… For Us Plutocrats – increasing inequality will eventually lead to the collapse of the sysem (at least so does teach the history). It is people – primarily the middle class – that are the source of the wealth of the society, they produce and also consume most. Thus it is necessary to support them …
- Why the U.S. Corporate World Became ‘A Bull Market for Corruption’ – Enron, GM, Goldman Sachs, … – we hear more and more the names of large corporations in the context of negligence and misues of their customers and investors. It seems that leadership (in the lead by example sense) has died out as well as the feeling of responsibility when one wields power over her customers/investors/markets. Instead, we have the me-first and money at any cost thinking. Organizations are designed to shield higher-ups from responsibility (meetings with no records…). High pay for short term gains, failure to punish high ranking people.
- (US) This is what happened when I drove my Mercedes to pick up food stamps – the experience of life in poverty after dropping down from $125k to $25k/year in two months due to childbirth, real estate market crash, and loss of a job. “Using the coupons was even worse. The stares, the faux concern, the pity, the outrage — I hated it. [..] That’s the funny thing about being poor. Everyone has an opinion on it, and everyone feels entitled to share. [..] Poverty is a circumstance, not a value judgment. I still have to remind myself sometimes that I was my harshest critic. That the judgment of the disadvantaged comes not just from conservative politicians and Internet trolls. It came from me, even as I was living it.“
- Isomorphic Clojure[Script], part I – enjoying all the benefits of Single-Page Apps while avoiding their drawbacks (SEO, slower page load, accessibility etc.) – a SPA that can be pre-rendered by the server. Using Om/React, JDK8 with the Nashorn JS engine, core.async, Sente (bi-dirrectional HTTP/WS communication over core.async) and Clojure in the JVM, ClojureScript in Nashorn in the JVM, and ClojureScript in the browser. Example app: Omelette.
- clj-crud: a relatively feature-complete example of a Clojure web (4/2014; GitHub) – using Component, Liberator (REST), Datascript + Quiescent (=> React.js), Enlive, Friend etc. including couple of unit-test and ui-test libraries
- Conclujon: Acceptance testing tool (α), Clojure reimplementation of Concordion, a beautifully simple ADD tool
- dynalint: human-friendly error messages during dev – Clojure typically provides little helpful and sometimes confusing error messages thrown from somewhere deep in the implementation, such as “Don’t know how to create ISeq from: java.lang.Long at clojure.lang.RT.seqFrom” while we want st. like “First argument to clojure.core/first must be seqable: 1 (instance of class java.lang.Long” – and that’s what Dynalint does. In the tradition of defensive programming, it adds checks and good error messages to Vars at runtime. You typically run it only during dev, triggering it from the REPL.
- Grimoire (Reid McKenzie) – a more up-to-date replacement for ClojureDocs
- Adam Bard’s Top Clojure Articles for beginners and intermediate Clojure devs – f.ex. Five Mistakes Clojure Newbies Make, Acceptable Error Handling in Clojure, Clojure Reducers for Mortals
- J. Wilk: Isolating External Dependencies in Clojure – a nice overview of the options and their pros and cons – with-redefs, alter-var-root, Midje (using alter-var-root in a more controlled manner), higher-order-functions (#1!) etc.
- philandstuff’s detailed notes from Euroclojure 2014
- NixOS (via @bodil) – a new interesting “purely functional” Linux distribution – system configuration is fully declarative (think of Puppet/Chef) and it is always trivial to roll back, you can have multiple versions of a package, users can install non-global SW
- InfluxDB – time series, metrics, and events DB that scales; contrary to Graphite it can store richer data than Graphite and its single value; additional highlights: authorization for individual data, roll-up/clean old data, https API. Written in Go.
Posted in General, Languages, Top links of month | Tagged: bigdata, clojure, ClojureScript, design, economics, human, linux, mobile, society, types | Comments Off
Posted by Jakub Holý on September 30, 2013
- Stuff The Internet Says On Scalability For September 13, 2013 – a collection of interesting performance related articles with summaries (via @_dagi)
- Can you copy a culture? The NUMMI story (audio/transcript) – how the GM factory with the worst workforce has been turned around via a good application of Toyota Production System – “a truly inspiring story of human potential and how systems can be designed to bring the best or worst of of people.” And how GM failed to learn from it and to copy Toyota’s culture.
- The Reactive Manifesto – why to write reactive SW – “Reactive applications represent a balanced approach to addressing a wide range of contemporary challenges in software development. Building on an event-driven, message-based foundation, they provide the tools needed to ensure scalability and resilience. On top of this they support rich, responsive user interactions. We expect that a rapidly increasing number of systems will follow this blueprint in the years ahead.“
- NYT: Eiji Toyoda, Promoter of the Toyota Way and Engineer of Its Growth, Dies at 100 – learn about the life of one of the founders of lean thinking
- Gojko Adzic: How we solved our #1 product management problem – valuable experience of false assumptions, learning from users, and a much helpful UI remake: even if you build a product to scratch your itch, you have to test it with real users
- Don’t use Hadoop – your data isn’t that big – a great post about the downside of Hadoop and that there are much better options (large disks, large RAM, Pandas/R/Postgres) for data up to few TBs. “In addition to being more difficult to code for, Hadoop will also nearly always be slower than the simpler alternatives.”
- Gartner On Big Data: Everyone’s Doing It, No One Knows Why – golf talk / hype -driven initiatives FTW! “According to a recent Gartner report, 64% of enterprises surveyed indicate that they’re deploying or planning Big Data projects. Yet even more acknowledge that they still don’t know what to do with Big Data.”
- What makes Spark exciting – why it might be a good replacement for Hive/Hadoop, based on experiences with H/H: “Hive has served us well for quite a while now. […] That said, it has gotten to the point where Hive is more frequently invoked in negative contexts (“damn it, Hive”) than positive. (Primarily due to being hard to test, hard to extend.)” “We had heard about Spark, but did not start trying it until being so impressed by the Spark presentation at AWS re:Invent [..] that we wanted to learn more. [..] Spark, either intentionally or serendipitously, addresses both of Hive’s primary shortcomings, and turns them into huge strengths. (Easy to test, extend.) [..] I find the codebase small and very easy to read, [..] –which is a nice consolation compared to the daunting codebases of Hadoop and Hive.” Cons.: Spark is only pre-1.0, the author hasn’t yet tested it heavily.
- 10 Ways to Make Your Office Fun To Work In – because we spend there plenty of our time so why not have a pleasant/cosy, inspiring environment? Some tips: plants, not-your-boring-enteprprise-look-and-feel, open it to the nature (I want this!), design it as home, not office, provide play space (I am too into work to want to play but having a resting place for a nap is st. I’d love).
- Book: The Architecture of Open Source Applications (via @rmz) – learn by studying architectures of existing systems – “In these two books, the authors of four dozen open source applications explain how their software is structured, and why. What are each program’s major components? How do they interact? And what did their builders learn during their development?“
- Book: Seven Concurrency Models in Seven Weeks: When Threads Unravel – “how to exploit different parallel architectures to improve your code’s performance, scalability, and resilience” – threads & locks, actors, FP + immutability/futures/promisses, Software Transactional Memory etc., GPU, MapReduce on clusters, … (intro) Personally, I would prefer from theory to practice approach and mention of CSP (-> Go’s channels, core.async) and more.
- Books Amazon CEO requires his top execs to read (The Effective Executive by Peter Drucker, The Innovator’s Solution by Clayton Christensen, The Goal by Eliyahu Goldratt)
- Stanford engineers build computer using carbon nanotube technology (via
- NYT: The Banality of Systemic Evil – a good article about human tendency to “obey the system” thus potentially causing evil – and thus the need to resist the system, as heroic individuals such as Snowden, Hammond, Schwartz, Manning. See the famous Eichmann in Jerusalem for how “doing your job” can create evil – “[..] what happens when people play their “proper” roles within a system, following prescribed conduct with respect to that system, while remaining blind to the moral consequences of what the system was doing — or at least compartmentalizing and ignoring those consequences.” (Tip: The book Moral Mazes explores the ethics of decision making within several corporate bureaucracies => mid-managers rules of life: (1) never go around your boss, (2) tell the boss what she wants to hear, (3) drop what she wants dropped, (4) anticipate what the boss wants so that she doesn’t need to act as a boss to get it, (5) do not report something the boss does not want reported, cover it up; the the job & keep your mouth shut.) “The bureaucracy was telling him [Snowden] to shut up and move on (in accord with the five rules in “Moral Mazes”), but Snowden felt that doing so was morally wrong.” “[..] there can be no expectation that the system will act morally of its own accord. Systems are optimized for their own survival and preventing the system from doing evil may well require breaking with organizational niceties, protocols or laws.“
- Fairphone – “A seriously cool smartphone that puts social values first” (likely the only one not built by poorly paid workers and creating too much ecological burden), for just €325. You can see detailed cost breakdown, list of suppliers, specs, and essentially everything. This is, in my opinion, super cool! Go and read the story!
- Amazonica – “A comprehensive Clojure client for the entire Amazon AWS api.”
- Talk Ritz, The Missing Clojure Tooling (40min, 9/2013) – thanks to this I finally understood how to use Ritz but it still seems not to work well, f.ex. setting a breakpoint always reported “Set 0 breakpoints” (lein ritz/middleware 0.7.0, nrepl-ritz.el 0.7.1); according to callen, debug-repl is simpler and nicer if you only care about local vars and evaluation. To try ritz: use M-x nrepl-ritz-jack-in, then M-x nrepl-ritz-break-on-exception, exec. f.ex. “(/ 1 0)”. In the poped-up buffer, t or enter to show frame locals, e to eval a code in the context of the frame etc. If you managed to trigger the debug buffer through a breakpoint, the actions lists would contain STEP etc. (See fun. nrepl-ritz-line-breakpoint)
- C. Grand’s spreadmap – “library to turn Excel spreadsheets in persistent reactive associative structures” => access content via map functions; changing a value updates formula cells using it
- Alembic Reloads your Leiningen project.clj Dependencies – add a dependency to your project.clj w/o needing to restart your REPL (just call (alembic.still/load-project), provided you have it in your lein dependencies). Limitations: cannot remove deps or change versions.
- Defeating stack overflows – techniques for transforming mutually recursive calls etc. into something that won’t blow the stack – “Priming the pump” (memoize subresults first), core.async
- Google Groups: Clean Architecture for Functional Programming – How do the Clean Architecture and the Clean Code best practices apply to FP (Clojure/Haskell)? Some points: OOP isn’t worse than FP, only people do class-oriented programming instead; OO better e.g. for UIs, combining them (func. core, imperative shell) can be sometimes best. Some clean arch. patterns are actually more like functions – “Interactors and Presenters, for example, do not maintain any state of their own. Even those objects that do imply some kind of state, such as entities and gateways, keep that state hidden behind boundaries and present a functional interface instead.“
- night-vision: Handy, super light weight debugging utility – add it to your lein profile and then call
(night-vision.goggles/introspect-ns! '<name of ns>) and it will print each entry/exit of a function within the scope of the namespace with the argument/return values
- Nil Punning (Or Null Pointers Considered Not So Bad) – a great post about why nil in Clojure is not bad contrary to Java’s null (because it is actually an object, you can call functions on it, treat it as false/empty list/map/set, most core functions work on it)
Posted by Jakub Holý on August 31, 2013
Sorry folks, this month it will be very brief. I have many more great stuff in the queue but haven’t managed to write it down yet. Next month will be heavy :-)
- Why Software Projects are Terrible and How Not To Fix Them – many teams are not ready to embrace new/better software practices, primarly for two reasons: 1) most of them are nonintuitive (f.ex. adding more people will slow dev down) and need to be sold through a high hierarchy of managament – but people/managers/organizations don’t really care, it takes years for good/bad practices to have an impact, which is not relevant “now.” 2) Businss objectives change too quickly and SW is blamed for not delivering. Based on evaluating many failed projects. Conclusion: Choose carefully people/organizations your work with. Avoid blame-driven ones. Quote on middle managers: “He has to put more developers on the project, call a meeting and yell at people, and other arbitrary bad ideas. Not because he thinks those will solve the problem. In fact, managers often do this in spite of the fact that they know it’s bad. Because that’s what will convince upper management that they’re doing their best.” “In the vast majority of failed projects I’ve been called to looked at, the managers have not read one book on software engineering.“
Data & Analytics
- Big Data: Kafka for uSwitch’s Event Pipeline – a better alternative to log files – use LinkedIn’s Kafka for messaging, have MR jobs to import latest messages into Hadoop/HDFS. The advantage of Kafka is that it persists the messages for a period of time so it is easy to batch-import and even re-import them. The uSwitch’s talk Users As Data explains the downsides of log files. LinkedIn’s Camus is a tool for importing messages from Kafka to HDFS.
- Realtime Analytics with Storm and Hadoop (at Twitter; presentation deck) – pre-aggregate some data into a read-only, random read DB such as ElephantDB, Voldemort, or Manhattan. For newer data use Storm and aggregated data in a read-write, big-data DB such as HBase, Riak, or Cassandra. For stuff that cannot be pre-aggregated you might use Storm’s Distributed RPC.
- The Unified Logging Infrastructure for Data Analytics at Twitter – a paper from late 2012 that presents “Twitter’s production logging infrastructure and its evolution from application-speciﬁc logging to a uni- ﬁed “client events” log format, where messages are captured in common, well-formatted, ﬂexible Thrift messages” – with the benefit of “s streamlined log collection and data analysis”.
- Development and Deployment at Facebook (Kent Beck et. al., 8/2013, 13p paper) – “More than one billion users log in to Facebook at least once a month to connect and share content with each other. Among other activities, these users upload over 2.5 billion content items every day. In this article we describe the development and deployment of the software that supports all this activity, focusing on the site’s primary codebase for the Web front-end.“
- One of the most valuable talks I’ve seen, in just 18 min: The Progress Principle – about the disengagement crisis and motivation at work by Teresa Amabile at TEDx Atlanta (via @thovden). Disengagement from work is increasing, at all age and salary levels, and leads to unhappy people, low productivity, huge financial losses. Based on analysing diaries of 12k participants, the single most important engaging and motivating factor is making progress in a meaningful work (including small wins). A culture of management by fear and punishment for failure creates disengagement and can crush even an innovative, profitable, praised company in a few years. Everybody, though especially the management, creates the culture through their everyday, small actions. If everybody focuses on catalysing progress and supporting their fellow humans through good and bad times, engagement and success will follow. Remove progress inhibitors, nourish the human spirit (acknowledge what we humans value, encourage people). Yet of the managers asked, very few knew of the significance of making progress (or, I can assume, of supporting people and making them happy(er) and the impact of our inner work life (perceptions, emotions, etc.) on our productivity and creativity). The study included two seemingly similar, successfull companies, one with great engagement, another with a new management that managed to destroy the engagement and thus eventually the company. Actions to take: catalyse progress, celebrate wins, encourage and support your colleagues.
- Wonderful Clojure Cheatsheet 1.5 with tooltips showing the doc and summary of information available at clojuredocs.org (other Clj versions), by Andy Fingerhut
- Chas Emerick’s Clojure type selection flowchart to help you decide whether to use a map, a record, reify, proxy, gen-class, or deftype. (Reify and proxy don’t produce a class but just an instance of an anonymous class; proxy can extend a base class, reify cannot. gen-class produces a class visible from Java and can extend Java classes. …)
- Docker.io – pack, ship and run any application (and its dependencies) as a lightweight container, i.e. essentially “a VM without the overhead of a VM,” using linux containers (chroot on steroids with resource limits via control groups) see reports of some uses such as Java app deployment, desktop virtualization, automatic app deployment in GitHub commit. Docker also supports evolving the containers over time, i.e. deploying new version, by pushing just diffs so it’s low-overhead. You can build a container (include files, SW, forward ports, …) using a Dockerfile. See dotScale 2013 – Solomon Hykes – Why we built Docker for an intro (20 min).
- Packer.io – tool for building pre-configured VM images for different platforms (EC2, VirtualBox, …), remotely similar to Netflix’s Aminator. See Immutable Servers With Packer and Puppet for an example use case.
- Ubuntu-build Vagrant boxes at cloud-images.ubuntu.com/vagrant/
- SlimerJS – PhantomJS-compatible headless browser engine based on Firefox/Gecko (well, it is not fully headless yet but that is planned; the main focus now is full compatibility with PhantomJS’ API) (Both work with CasperJS for navigational steps/testing.)
- localtunnel – instantly show locally running webapp/server to the rest of the world (gem install localtunnel, localtunnel <port to share>, => share the url returned, e.g. http://xyz.localtunnel.com) – I haven’t tried it but it looks simple and very convenient
- Logstash + Kibana (via @mortenberg): take control of your logs – while Logstash can collect (from multiple servers/services), parse (over 100 built-in patterns), store, index, search your logs, Kibana is a web interface to seach them, view them in realtime (based on a query) etc. See this Logstash slides (9/2012) and an overview of Kibana’s powers. PS: Logstash can also compute metrics and send them to graphite etc. It is typically used with ElasticSearch.
- ncdu is an interactive, command-like disk usage browser that shows a list of directories sorted by size shown in human-friendly units, you can navigate with arrows and enter and i to show the current dir/file info, d to delete it, q to quit; check out this article about ncdu with screenshots and ncdu man page. Install via Apt etc., run f.ex. with ncdu -x / .
- vagrant-cachier – Vagrant plugin for caching apt/yum/.. packages locally, thus speeding up destroy+up
Posted by Jakub Holý on June 30, 2013
Agile, process, SW dev, people etc.
- Real Options—a Mindset – an intro into the Real Options approach, which has been quite a hot topic and a transformational way of thinking for a number of inspiring people (Dan North, Liz Keogh etc.). “Real Options help us to better make decisions and commitments with three simple principles: Options have value. Options expire. Never commit early unless you know why.” We can “pay” to keep our options open longer, i.e. to avoid commiting prematurely.
- Demystifying the CHAOS report’s claim of ~ 1/2 features being unused: the Standish Group’s CHAOS report has been often quoted for its “finding” that a large percentage of features in applications is never/rarely used. However this claim seems to have never been confirmed, their “research” is reportedly not very scientific and not publicly available for scrutiny. Critique by Laurent Bossavit (2013), Jorge Aranda’s Standish, the CHAOS report, and science. Thx to @smalltalk80 for pointing this out! However there is one research, Online Experimentation at Microsoft, that supports the claim, in a different context but the same problem applies to features: “Evaluating well-designed and executed experiments that were designed to improve a key metric, only about one-third were successful at improving the key metric!”
- Why Yammer believes the traditional engineering organizational structure is dead – small teams, small projects (2-10 people, 2-10 weeks), no separation into front/middle tier/backend team (=> communication, design obstacle); have instead people specializing in these areas and construct feature teams from them based on the actual needs; engineers, not managers do eng. decisions; all aligned via focusing on the same 3 key metrics. Small projects => constant sense of urgency (and excitement): Often very long projects cause engineers to lose track of the end goal. Think of it in terms of hiking: start fresh & excited, get tired and losing track of the goal, excited again at the end => cut out the middle part, keep them in the exciting state where they can measure progress and see it visually; it’s the only way to maintain urgency and morale. Focus: people alwasy work only at one (short) project at a time (there are special bug-fixing teams for maintenance tasks with people rotating in&out).
- Agile development is more culture than process – Why thinking of agile as culture and not just process explains resistance and difficulty in teaching and learning the approach – and should be taught so => 1. Underscore agile values that motivate practice; 2. Identify organization values that compete with agile values, conflict of values; 3. Be sensitive to culture shock.
- Mark Zuckerberg’s Letter to Investors: ‘The Hacker Way’ (quite long, you might want to read only “The Hacker Way” part at the end) – about Facebook’s “unique culture and management approach” – “Hackers believe that something can always be better, and that nothing is ever complete.” “Hackers try to build the best services over the long term by quickly releasing and learning from smaller iterations rather than trying to get everything right all at once.” “Instead of debating for days [..], hackers would rather just prototype something and see what works.” “Hacker culture is also extremely open and meritocratic.” “Many of our most successful products came out of hackathons, [..].” <=> five core values: Focus on Impact (focus on solving the most important problems, be good at finding the biggest problems to work on); Move Fast (“[..] if you never break anything, you’re probably not moving fast enough.”); Be Bold (“Building great things means taking risks.”); Be Open (=> effort to make as much info as possible visible to all); Build Social Value (“[..] Facebook exists to make the world more open and connected, and not just to build a company. “)
- Dave Nicolett: I know how to tie my shoes – on the difficulty of convincing people to try unfamiliar software development techniques – “People change the way they operate when they are experiencing some sort of inconvenience or negative feedback. As long as things are going along reasonably well, people don’t go out of their way to change the way they work.” (with few exceptions) You can learn to tie your shoes in a split second, but why to invest the effort? You’d need to set aside assumptions, suppress habits, practice. You can argument there are many inconveniences (bugs, criticism for slow delivery, …) but “Unfortunately, that’s all pretty normal, and most people in the software field are accustomed to it. They don’t see it as a problem that calls for them to change their practices. Most of them probably have a hard time visualizing a different reality.” => Maybe that’s the reason there’s been no satisfactory answer to the question of how to convince people to adopt different practices. We shouldn’t be trying to convince people to do anything. We should be helping people solve their problems and achieve their goals. If they are satisfied with the outcomes they achieve using their current methods, then there is no problem to solve.
- Kent Beck: Pace of Progress = Pace of Feedback – ‘”The pace of my progress is completely constrained by the pace of my feedback”. If I want to go faster, it’s hard to achieve by going faster. I can almost always optimize my feedback loop, though.’ “The second lesson from this episode is that it’s not just the duration of the feedback loop that matters, it’s also the quality. All week I was working in tiny little iterations. Without producing useful information, though, those iterations could be as small or as large as I liked, I was still just going to spin my wheels.” => “The next time I seem to be going slow, I’m going to look at my whole feedback loop–duration, quality and my ability to respond to the information.“
- What Google Has Learned About How to Hire People – interview results have no relation to actual performance on the job: “We looked at tens of thousands of interviews, and everyone who had done the interviews and what they scored the candidate, and how that person ultimately performed in their job. We found zero relationship. It’s a complete random mess.” “Instead, what works well are structured behavioral interviews, where you have a consistent rubric for how you assess people, [..]” ‘Behavioral interviewing also works — where you’re not giving someone a hypothetical, but you’re starting with a question like, “Give me an example of a time when you solved an analytically difficult problem.”’ Link to an interesting book, Hiring Geeks That Fit.
Cool tech stuff
- The Elixir language – Clojure + Ruby + Erlang – a functional meta-programming aware language built on top of the Erlang VM; a dynamic language with flexible syntax with macros support that leverages Erlang’s abilities to build concurrent, distributed, fault-tolerant applications with hot code upgrades. First-class support for pattern matching, polymorphism via protocols, etc. (via @bodil)
- Random Testing seems to be gaining popularity and looks very interesting; at NDC Oslo, John Hughes has presented how QuickCheck, which generates random sequences of API calls, has been successfully used to find bugs in the Riak DB and a file system that a human would never think of, and Stuart Halloway has presented simulation testing with Simulant, which runs predefined actions according to a probabilistic model (e.g. 100 traders, each having 1h mean time between trades and mean traded amount 100, the test runs for 4 simulated hours). Something worth exploring!
- Dmytro Navrotskyy’s collection of Frontend Development resources and learning materials for tools (grunt, unused css detection,..), best practices (Atomic Design, …), JS/CSS frameworks, typography, animation, visualization, useful on-line services, and many more (via Herman Schistad)
- The Secret To 10 Million Concurrent Connections – The Kernel Is The Problem, Not The Solution: To have really fast SW, you need to implement your own core services (FS, net driver (packet handling), thread scheduling, ..) tuned for your app. You need to be aware of the clock-time cost of cache misses, memory access etc.. Custom solutions are times faster than what the general OS kernel can offer. => “data plane oriented system” Core areas and solutions for them: packet scalability, multi-core scalability (locks are expensive), memory scalability.
- M. Fowler: EmbeddedDocument – a pattern for working with JSON flowing in/out of our services (REST <-> JSON-friendly DB) without unnecessary conversions but with good encapsulation; naive approach: json -> object graph -> (processing) -> json; “In many of these situtiations a better way to proceed is to keep the data in a JSONish form, but still wrap it with objects to coordinate manipulation.” – use a lib to parse the JSON into a generic structure (e.g. a structure of lists, and maps/dicts) and store in a field of an object defining methods that encapsulate it – f.ex. for an Order we could have a method returning the customer and another computing the cost, accessing the underlying generic structure. The user of the wrapper object doesn’t need to know/care about the underlying structure. “The sweet spot for an embedded document is when you’re providing the document in the same form that you get it from the data store, but still want to do some manipulation of that data. [..] The order object needs only a constructor and a method to return its JSON representaiton. On the other hand as you do more work on the data – more server side logic, transforming into different representations – then it’s worth considering whether it’s easier to turn the data into an object graph.”
- ThoughtWorks’ Approach To Big Data Analytics – an inspiring, brief read. Some really good points such as “It’s not about Data. It’s about Insight and Impact” => “focus on the questions you’d love to answer for your business” => “changing big data from a technological problem to a business solution.” Also “The value of data is only realised through insight. And insight is useless until it’s turned into action.” Measure the value you gain at each step. See Introducing Agile Analytics: A Value-Driven Approach to Business Intelligence and Data Warehousing by Ken Collier
- Wired.com, Nassim Taleb: Beware the Big Errors of ‘Big Data’ – in big data, noise has much stronger effect and in a large enough dataset we will always find spurious (i.e. false) relationships => beware! “Well, if I generate (by simulation) a set of 200 variables — completely random and totally unrelated to each other — with about 1,000 data points for each, then it would be near impossible not to find in it a certain number of “significant” correlations of sorts. But these correlations would be entirely spurious.”
- A Taste of Salt: Like Puppet, Except It Doesn’t Suck – a deescription of Salt and the tools around by an enthusiastic user with deep experience with Puppet. Highlights: Light-weight communication over ZeroMQ, very active community, simplicity, configuration is YAML, Salt-cloud can spin instances in EC2/Openstack/…, Salt-virt does the same for virtual machines (KVM/Xen/…), Salt-vagrant, Salt-monitor (work in progess) can ask all the server for their stats. “Having stood up a number of different configuration management systems across a wide variety of environments, I’ve yet to find a solution that’s as rapid to deploy, simple to scale, or as well architected as Salt.”
- Trash Your Servers and Burn Your Code: Immutable Infrastructure and Disposable Components – leveraging the lectures of PF to have a stable infrastructure – instead of updating servers, throw them away and create a new one from scratch (requires virtualization/cloud); this is something that Netlfix is doing and also Comoyo is moving towards
- Robin Ward: AngularJS vs Ember – a nice overview of the different approaches of the two; the author is strongly pro-Ember, claiming that AngularJS is much closer to low-end libraris like Backbone/Knockout and that you will often need the additional features of Ember. The comments provide the right countrweight to the biased post and form thus a good whole together.
- Scala Productivity. A Survey of the Community – people (that asnwered) seem to be productive with Scala right from the start
- After Your Job Is Gone – an interesting essay on the future, which, according to the author, we can already see happening, when technology will take away most of our work and we will not need to work all day. Not very optimistic, though (the author predicts few reach and many poor people).
- Clojure Cup 2013, Sept 28-29 – create something cool with Clojure/ClojureScript within 48h and perhaps win a price! #fun
- Clojure use in the industry – examples at an e-mail forum – Netflix, Puppet Labs (e.g. PuppetDB), UBS (talk), Deutsche Bank (talk, some details), Citigroup (reportedly “the largest private sector deployment of Clojure to date,” 11/2012), getprismatic.com (with frontend moving to ClojureScript; -> Why Prismatic Goes Faster With Clojure), Roomkey.com (details in a Relevance podcast), MastodonC.com (big data), Trend Micro, Walmart, beanstalkapp.com, ReadyForZero.com (50kLoC), www.cognician.com (20kLoC), World Singles (13kLoC) and more… (another similar thread)
- Stuart Sierra’s My Clojure Workflow, Reloaded (6/2013) – mainly about reloading changes into REPL, working around things that are not reloaded/left over => restart the app from scratch after significant changes => the app as a transient object => no global state, careful management of resources, :dev profile with :source-paths to a dir with user.clj (autoloaded by repl, pre-loading useful stuff) and dev util deps
- Adam Bard’s walk-through useful Clojure libs – f.ex. clojure.[data.[csv xml json] inspector java.shell java.browse xml], tools.logging, clojure.core.[match logic typed contracts …]
- Juxt.pro: Jon Pither’s and Malcolm Sparks’ “network of experienced IT professionals who specialise in the Clojure programming language,” providing training, consulting, talks
- Anthony Grimes: The Clojure Community and Me (2011) – an exciting insight into the embracing and supportive Clojure community
- In Clojure-based Machine Learning: “Our backend is 99.4% coded in Clojure, and 66% of the team [of 3] had never programmed seriously in any Lisp, let alone Haskell or Prolog (heck, not even I (the remaining 33%) had actually tried anything non-mainstream for real in a big project!) Maybe some Ruby, and lots and lots of Java and C and C++. But they accepted the challenge after reading around and learning the basics, and 3 months later you couldn’t take Clojure from their prying hands.”
- J. Pither: TDD and Clojure – “If you were to create a shopping list of things you really want for your development experience then what would you put at the top?” => 1. rapid feedback on changes, 2. REPL (place to explore and to play with your code <=> TDD), 3. FP and Immutability (“FP and dynamic languages lead to a lot less code. There’s less ceremony, less modeling. Because you’re managing less code you do less large scale refactorings.” => TDD needed less), 4. Regression Tests (“It’s my current opinion that what you get left out of TDD once you have amazingly fast feedback and a REPL is regression testing.”)
- More beautiful and colorful git log, by Filipe Kiss (via @lcdutoit) You may also want to have a look at tig, which is a text-based UI for git with the default view similar to git log.
- jq (via lcdutoit) – sed/awk/grep for JSON – slice, filter, map, transform structured data
- SemanticMerge (Windows) – a Java-aware merge tool – free beta (I haven’t tried it)
- JetLang – a high performance java threading library for in-memory messaging, based upon Retlang (via @tastapod, used likely in a trading SW).
Agile [is] in NOT a process — it’s a philosophy.
- Joe Wroblewski in a comment to a blog post
Teach culture first, then process and techniques
- Jeff Patton in Agile development is more culture than process
If a candidate came telling me that s/he wanted to program only in, say, Java because that’s what s/he knows best and that s/he doesn’t really feel prepared or interested in learning and using, say, Clojure (or any other language, really), I wouldn’t hire her/him in a million years, no matter what language my project were using, and no matter how many thousands of candidates like this one I had at my disposal.
- José Antonio Ortega in Clojure-based Machine Learning
We shouldn’t be trying to convince people to do anything. We should be helping people solve their problems and achieve their goals. If they are satisfied with the outcomes they achieve using their current methods, then there is no problem to solve.
- Dave Nicolett in I know how to tie my shoes
Posted in General, Languages, Testing, Tools, Top links of month | Tagged: bigdata, clojure, DevOps, fp, frontend, Git, java, lean, performance, simulation, waste | Comments Off
Posted by Jakub Holý on April 30, 2013
The top top article
How To Survive a Ground-Up Rewrite Without Losing Your Sanity (recommended by Kent Beck) – sometimes you need to actually rewrite an important part of a system; here we learn about two such rewrites, one which went well and one that failed badly – and what are the important differences.
The pain of a rewrite: “it’s [a major rewrite] going to take insanely longer than you expect” – because: “there’s this endless series of weird crap encoded in the data in surprising ways” and it takes days to convert them, “It’s brutally hard to reduce scope” (you cannot drop features, edge cases), “There turn out to be these other system that use ‘your’ data”.
To succeed you need: 1) Determine clear business-visible wins to justify the effort that will be much higher than expected and to know when to give up / what to postpone; 2) Do it extremely incrementally (<-> Succession) – break it into a series of small, safe steps, each generating a business value and learning of its own thus enabling early and frequent economical tradeoffs (stop, shift priorities, …) – ex.: rewrite a single reports, migrate its data, switch customers to it, go on to the next one – complete slice of functionality => a more realistic estimate soon => reprioritisation; incrementalism requires you to be able to write data both to the old and new system, which is hard but always pays off: “Here’s what I’m going to say: always insert that dual-write layer. Always. It’s a minor, generally somewhat fixed cost that buys you an incredible amount of insurance.” 3) “Abandoning the Project Should Always Be on the Table” (<- known biz value, better estimate based on early feedback).
Some Specific Tactics: Shrink Ray FTW (a graph of how much has been already replaced => motivation), Engineer The Living Hell Out Of Your Migration Scripts (tests, robustness, error handling, restartability), If Your Data Doesn’t Look Weird, You’re Not Looking Hard Enough.
Methodology, agile, lean
- M. Fowler: The New Methodology – a good description of the rise of Agile, the motivation for it, the various Agile methodologies (XP, Lean, Scrum etc.) and what is required to be able to apply an agile approach. Main points: Agile is adaptive (vs. predictive) and relies heavily on people and their judgement and skills (vs. treating them as same, replacable units) – which also leads to the need of leadership instead of (command&control) management. Discusses unpredictability of requirements and scope, foolishness of separating design and implementation, difficulty of measurement of SW development, continuous improvement etc. Quotes: “However letting go of predictability doesn’t mean you have to revert to uncontrollable chaos. Instead you need a process that can give you control over an unpredictability. That’s what adaptivity is all about.”
- The Toyota concept of ‘respect for people’ – many state that they respect their workers but fail to really understand what it means; it is not about freedom of act, it is about a mutual respect, leveraging the strengths of each other: worker’s experience and insight and manager’s broader overview, as demonstrated by the problem-solving dialog and challenges (problem – root cause – solution – measure of success, the manager challenging the worker’s answers). Also a nice example how the evaluation of individual performance leads to a much worse system and high turnover compared to a whole-oriented company.
- Fixed Bid Agile Without Cognitive Dissonance – a refreshing take on fixed-scope projects and Agile; yes, they are bad but sometimes the client has no other choice so what best we can make out of it? The core advice: Agree “a pragmatic change management protocol (along with a contingency built into the pricing)” (push for lower initial requirements granularity, customer involvement, flexibility of functionality) => “you can gain significant agile benefits for clients who wouldn’t otherwise accept them”.
- Agile Atlas: Scrum – a good description of Scrum and its values, roles, artifacts, and activities
Learning, psychology, estimates
- How Developers Stop Learning: Rise of the Expert Beginner – sometimes you meet people with experience-indicating titles that are actually little competent, perhaps leading incompetent IT departments. Why? They, unchallenged by competent peers or broader IT community, came to believe that they are “experts” while actually being only little more advanced beginners, better than their beginner colleagues but still lacking any understanding of the big picture and the knowledge of what they do not know, trapped in the “unconscious incompetence” stage. The post explains this in a more detail and is followed up an explanation how it can lead to the rise of a mediocre SW group in “How Software Groups Rot: Legacy of the Expert Beginner“.
- Coding, Fast and Slow: Developers and the Psychology of Overconfidence (via @peterskeide) – why are we so bad at estimating (inherent complexity of SW vs. our overconfidence) and why it cannot be fixed. We can learn to somehow estimate tasks of few hours length (less complex, plenty of practice opportunities). The question is: “how you can your dev team generate a ton of value, even though you can not make meaningful long-term estimates?”
- Cognitive Overhead, Or Why Your Product Isn’t As Simple As You Think (via @JiriJerabek) – to make apps more accessible to users, we try to make them simple – but “simple” might be different from what you expect. The important thing is not less steps, less features, less elements, but lower cognitive overhead, i.e. “how many logical connections or jumps your brain has to make in order to understand or contextualize the thing you’re looking at.” Good examples of unexpectadly high / pleasantly low cognitive overhead, some tips, even suprising ones such as make people do more (to be more involved in the process – e.g. bump their phones), slow down your product.
- Economies of Scala – a case for using Scala over Java, supported by data: many capable developers want to use it but there are few opportunities for them – and getting developers is one of the main challenges.
- A canonical Repository test – a nice standard way to test a “DAO”; highlights: use of FEST assert 2 for clean and nice checks, no unimportant details in the test (f.ex. details of the test data hidden in randomPerson() and randomOder(Person)).
- How To Think Like An Engineer – some nice ideas such as: “Build A Simple First Version: With People, Not Code” – “Technology is not always the best solution, because technology is not always the simplest solution.”, i.e. don’t automate everything from the start (examples from Netflix, Amazon); “Rather than trying to do everything at once, break down the functions of your company into smaller goals.” – and focus at one at a time
- Economies Of Scale As A Service (do not mix up with Scala! :-))- an interesting description of the trend away from ownership to the rental of important resources (servers, manufacturing capabilities, personal cars, …) and the resulting changes in the society, business, and industry
- Troy Hunt: Our password hashing has no clothes (or the much shorter though biased How To Safely Store A Password) – MD5 and SHA are not safe enough due to brute-force attack enabled by GPUs, irrespective key size; it’s crucial to use hashing algorithms designed for passwords (and thus sufficiently slow) – f.ex. bcrypt, or PBKDF2 or the newer scrypt.
- Everything about Java 8 – a well-made summary of what should come in Java 8, based on the current state, discussing the finer points: static and default (non-static, overridable) methods on interfaces, lambdas (do I need to mentione that?!) and method references (String::valueOf, Object::toString, myVar::toString, ArrayList::new); good discussion of the various use cases and limitations of lambdas (capturing x non-c., ..); java.util.stream for functional operations on value streams (filter, map, reduce etc.); java.time inspired by Joda, more concurrency utilities (e.g. CompletableFuture for chaining futures); String.join (finally!), Optional ~ Scala’s Option & more; yummy!
- How To Keep Your Best Programmers – what motivates capable programmers to stay/leave? The author lists some common reasons and concludes that, ultimately, all are linked to the desire for autonomy, mastery, or purpose. However he goes further and proposes that, to keep talented devs, you must offer them an appealing narrative (regarding their actions and a result, related to autonomy/mastery/purpose) and reaffirm/update it frequently; ex.: “With the work that we’re giving you over the next few months, you’re going to become the foremost NoSQL expert in our organization.” “At any point, both you and the developers on your team should know their narratives.” – so that they will be “constant points of job satisfaction and purpose.”
- Clojure Data Analysis Cookbook review – “The book provides a collection of recipes for accomplishing common tasks associated with analyzing different types of data sets. It starts out by showing how to read data from a variety of sources such as JSON, CSV, and JDBC. [..] how to sanitize the collected data and sample large data sets. [..] a number of different strategies for processing it.” How to present them with ClojureScript and NVD3 (D3.js components). “Some of the highlights include using the Clojure STM, parallel processing of the data, including useful tricks for partitioning, using reducers, and distributed processing with Hadoop and Casalog.”
once again, trying to do it *and* do it right was too much all at once, resulting in little progress and little learning.
- Kent Beck’s tweet 2013-04-16
A true agile development process can be recognized by its continual evolution:
A project that begins using an adaptive process won’t have the same process a year later. Over time, the team will find what works for them, and alter the process to fit.
- Martin Fowler in The New Methodology
Posted in General, SW development, Testing, Top links of month | Tagged: agile, bigdata, clojure, human, learning, scrum | Comments Off
Posted by Jakub Holý on January 31, 2013
- Dustin Marx: Significant Software Development Developments of 2012 – Groovy 2.0 with static typing, rise of Git[Hub], NoSQL, mobile development (iOS etc.), Scala and Typesafe stack 2.0, big data, HTML5, security (Java issues etc.), cloud, DevOps.
- 20 Kick-ass programming quotes – including Bill Gates’ “Measuring programming progress by lines of code is like measuring aircraft building progress by weight.”, B.W. Kernighan’s “Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.”, Martin Golding’s “Always code as if the guy who ends up maintaining your code will be a violent psychopath who knows where you live.” (my favorite)
- How to Have a Year that Matters (via @gbrindusa) – do you want to just survive and collect possessions or do you want to make a difference? Some questions everybody should pose to him/herself.
- Expression Language Injection – security defect in applications using JSP EL that can sometimes leads to double evaluation of the expressions and thus makes it possible to execute data supplied by the user in request parameters etc. as expressions, affects e.g. unpatched Spring 2.x and 3.
- HN discussion about Scala 2.10 – compilation speed and whether it matters, comparison of the speed and type system with Haskell and OCaml, problems with incremental compilation (dependency cycles, fragile base class), some speed up tips such as factoring out subprojects, the pros and cons of implicits etc.
- Blog Mechanical Sympathy – interesting posts and performance tests regarding “writing software which works in harmony with the underlying hardware to gain great performance” such as Memory Access Patterns Are Important and Compact Off-Heap Structures/Tuples In Java.
- Neal Ford: Functional thinking: Why functional programming is on the rise – Why you should care about functional programming, even if you don’t plan to change languages any time soon – N. Ford explains the advantages of FP and why FP concepts are spreading into other languages (higher abstractions enabling focus on the results over steps and ceding control to the language, more reusability on a finer level (higher-order functions etc.), few generic data structures with many operations -> better composability, “new” and different tool such as lazy collections, shaping the language towards the problem instead of vice versa, aligning with trends such as immutability)
- Neal Ford: Java.next: The Java.next languages Leveraging Groovy, Scala, and Clojure in an increasingly polyglot world – a comparison of these languages with focus on what they are [not] suitable for, exploration of their paradigms (static vs. dynamic typing, imperative vs. functional)
- How to Completely Fail at BDD – a story of an enthusiastic developer who tried to make everyone’s life better by introducing automated BDD tests and failed due to differences in culture (and inability to change thinking from the traditional testing), a surprising lack of interest in the tool and learning how to write good tests: “Culturally, my current team just isn’t ready or interested in something like this.” Morale: It is hard to change people, good ideas are not enough.
- M. Feathers: Refactoring is Sloppy – refactoring is often prioritized out of regular development and refactoring sprints/stories aren’t popular due to past failures etc. An counter-intuitive way to get refactoring in is to imagine, during planning, what the code would need to be like to make it easy to implement a story. Then create a task for making it so before the story itself and assign it to somebody else then the story (to force a degree of scrutiny and communication). “Like anything else in process, this is medicine. It’s not meant to be ‘the way that people do things for all time’ [..]” – i.e. intended for use when you can’t fit refactoring in otherwise. It may also make the cost of the current bad code more visible. Read also the commits (f.ex. the mikado method case).
- Cyber-dojo: A great way to practice TDD together. Compare your read-green cycle and development over time with other teams. Purposefully minimalistic editor, a number of prepared tdd tasks.
- On the Dark Side of “Craftsmanship” – an interesting and provoking article. Some developers, the software labouers, want to get work done and go home, they haven’t the motivation and energy to continualy spend time improving themselves. There is nothing wrong with that and we shouldn’t disparge them because of that. We shouldn’t divide people into craftsmen and the bad ones. A summary of and response to the varied reactions follows up in More on “Craftsmanship”. The author is right that we can’t expect everybody to spend nights improving her/his programming skills. Still they should not produce code of poor quality (with few exceptions) since maintaining such code costs a lot. There should be time for enough quality in a 9-5 day and people should be provided with enough guidance and education to be able to write decent code. (Though I’m not sure how feasible it is, how much effort it takes to become an acceptable developer.) Does the increased cost of writing (an learning to write) good code overweight the cost of working with bad code? That is an eternal discussion.
Cloud, web, big data etc.
- Whom the Gods Would Destroy, They First Give Real-time Analytics (via Leon) – a very reasonable argument against real-time analytics: yes, we want real-time operational metrics but “analytics” only makes sense on a sensible amount of data (for the sake of statistical significance etc.) RT analytics could easily provide misguided results.
CAP Twelve Years Later: How the “Rules” Have Changed (tl;dr, via @_dagi) – an in-depth discussion of the CAP theorem and the simplification (2 out of 3) that it makes; there are many more nuances. By Eric Brewer, a professor of computer science at the University of California, Berkeley, and vice president of infrastructure at Google.
- ROCA: Resource-oriented Client Architecture – “A collection of simple recommendations for decent Web application frontends.” Server-side: true REST, no session state, working back/refresh etc. Client: semantic HTML independent of layout, progressive enhancement (usable with older browsers), usable without JS (all logic on the server) etc. Certainly not suitable for all types of apps but worthwile to consider the principles and compare them with your needs.
- Vaurien, the Chaos TCP Proxy (via @bsvingen) – an extensible proxy that you can control from your tests to simulate network failure or problems such as delays on 20% of the requests; great for testing how an application behaves when facing failures or difficulties with its dependencies. It supports the protocols tcp, http, redis, memcache.
- Wvanbergen’s request-log-analyzer for Apache, MySQL, PostgreSQL, Rails and more (via Zarko) – generates a performance report from a supported access log to point out requests that might need optimizing
- Working Effectively With iTerm2 (Mac) – good tips in the body and comments
A very good (though not very scientific) definition of project success applicable for distinguishing truly agile from process-driven projects:
[..] a project is successful if:
- Something was delivered and put to use
- The project members, sponsors and users are basically happy with the outcome of the project
- Johannes Brodwall in “How do we become Agile?” and why it doesn’t matter, inspired by Alistair Cockburn
(Notice there isn’t a single word about being “on time and budget”.)
Posted in General, Languages, Testing, Tools, Top links of month | Tagged: bigdata, clojure, cloud, fun, human, java, performance, SbE, scala, security, tdd, Testing | Comments Off
Posted by Jakub Holý on December 31, 2012
- Kent Beck: When Worse Is Better: Incrementally Escaping Local Maxima – Kent reintroduces his Sprinting Centipede strategy (“reduce the cost of each change as much as possible so as to enable small changes to be chained together nearly continuously” => “From the outside it is clear that big changes are happening, even though from the inside it’s clear that no individual change is large or risky.”) and advices how to deal with situations where improvements have reached a local maxima by making the design temporarily intentionally worse (f.ex. by inlining all the ugly methods or writing to both the old and the new data store); strongly recommended
- Related: Efficient Incremental Change – transmute risk into time by doing small, safe steps, then optimize your ability to make these steps quickly and thus being able to achieve large changes
- Researchers: It is not profitable to outsource development – the Scandinavian research organisation SINTEF ICT has studied the effects of outsourcing and discovered that often it is more expensive than in-country development due to hidden costs caused by worse communication and cultural differences (f.ex. Indians tend not to ask questions and work based on their, often incomplete, understanding) and very high people turn-over; even after the true cost is discovered, companies irrationally stay there. However it is possible to succeed, in some cases.
- Bjørn Borud: Tractor pulling and software engineering – very valuable and pragmatic advices on producing good software (i.e. avoiding accumulating so much crap that the software just stops progressing). Don’t think only about the happy path. Simplify. Write for other developers, i.e. avoid too “smart” solutions, test & document, dp actually think about design and its implication w.r.t performance etc. Awake the scientist in you: “Do things because you know they work, not because it happens to be the hip thing to do.”
(Note: I see the good intention behind “design for the weakest programmer you can think of” but plase don’t take it too far! Software should be primarily simple, not necessarily easy.
- Know your feedback loop – why and how to optimize it – to succeed, we need to learn faster; the only way to do that is to optimize our feedback loops, i.e. shorten the path our assumptions travel before they are (in)validated, whether about our code, business functionality, or the whole project idea. Conscise, valuable.
- Code quality is the least important reason to pair program – the author argues, based on his experience, other benefits of pair programming are more important than code quality: “[..] the most important reasons why we pair: it contributes to an amazing company culture, it’s the best way to bring new developers up to speed, and it provides a great way to share knowledge across the development team.”
- You Can’t Refactor Your Way Out of Every Problem – refactoring can’t help you if the design is fundamentally wrong, you need to rewrite it; know when it can or cannot help and act accordingly (related to how much design is needed upfront since some design decision cannot be reverted/improved upon)
- Josh Bloch: Java – the good, bad and ugly parts (video, 15 min); summary: right design decisions (VM, GC, threads, dynamic linking, OOP, static typing, exceptions, …), some bad details (signed byte, lossy long-> double, == doesn’t cal .equals, ability to call overriden methods from constructors, …); Mr. Bloch has also given a longer talk examining the evolution of Java from 1.0 to 1.7 in The Evolution of Java: Past, Present, and Future.
- True Scala complexity – a thoughtful criticism of the complexity of Scala, based on code samples; “[it is true that] Scala is a language with a smaller number of orthogonal features than found in many other languages. […] However, the problem is that each feature has substantial depth, intersecting in numerous ways that are riddled with nuances, limitations, exceptions and rules to remember. It doesn’t take long to bump into these edges, of which there are many.”; however, its possible to avoid many of the problems mentioned by resorting to less smart, more clumsy and verbose Java-like code; also, the author still likes Scala.
- Scala or Java? Exploring myths and facts (3/2012) – a balanced view of Scala’s strengths and weaknesses; “[..] the same features that makes Scala expressive can also lead to performance problems and complexity. This article details where this balance needs to be considered.” Topics: productivity, complexity, concurrency support, language extensibility, Java interoperability, quality of tooling, speed, backward compatibility. Plenty of useful links.
Big data & Cloud:
- Dean Wampler’s slides from Beyond Map Reduce – 1) Hadoop Map Reduce is the EJB 2 of big data but there are better APIs such as Cascading with Scala/Clojure wrappers; there are also “alternative” solutions like Spark and Storm; 2) functional/relational programming with simple data structures (lists, sets, maps etc.) is much more suitable for big data than OOP (for we do mostly stateless data transformations)
- Apache HBase vs Apache Cassandra – comparison sheet – if you want to decide between the two
- Optimizing MongoDB on AWS – 20 min talk about the current state of the art. Simplicity: Mongo AMIs by 10gen, Cloudformation template etc. Stability & perf.: new storage options – EBS with provisioned IOPS volumes (high I/O) + EBS Optimized Instances (dedicated throughput to EBS), High IO instances (hi1.4xlarge – SSD)); comparison of throughput (number of operations, MBs) of these storages; tips for filesystem config. Scalability: scale horizontally and vertically, shrink as needed.
- Getting Real About Distributed System Reliability by Jay Kreps, the author of the Voldemort DB: distributed software is NOT somehow innately reliable; a common mistake is to consider only probability of independent failures but failures typically are dependent (e.g. network problems affect the whole data center, not a single machine); the theoretical reliability “[..] is an upper bound on reliability but one that you could never, never approach in practice”; “For example Google has a fantastic paper that gives empirical numbers on system failures in Bigtable and GFS and reports empirical data on groups of failures that show rates several orders of magnitude higher than the independence assumption would predict. This is what one of the best system and operations teams in the world can get: your numbers may be far worse.” The new systems are far less mature (=> mor bugs, worse monitoring, less experience) and thus less reliable (it takes a decade for a FS to become mature, likely similar here). Distributed systems are of course more complex to configure and operate. “I have come around to the view that the real core difficulty of these systems is operations, not architecture or design.” Some nice examples of failures.
- Talks To Help You Become A Better Front-End Engineer In 2013 (tl;dr) – topics such as mobile web development, modern web devel. workflow, current/upcoming featrues of CSS3, ECMAScript 6, CSS preprocessors (LESS etc.), how to write maintainable JS, modular CSS, responsive design, JS debugging, offline webapps, CSS profiling and speed tracer, JS testing
- On Being A Senior Engineer – valuable insights into what makes an engineer “senior” (i.e. mature; from the field of web operations but applies to IT in general): mature engineers seek out constructive criticism of their designs, understand the non-technical areas of how they are perceived (=> assertive, nice to work with etc.), understand that not all of their projects are filled with rockstar-on-stage work, anticipate the impact of their code (on resource usage, others’ ability to understand & extend it etc.), lift the skills and expertise of those around them, make their trade-offs explicit when making decisions, do not practice “Cover Your Ass Engineering,” are able to view the project from another person’s (stakeholder’s) perspective, are aware of cognitive biases (such as the Planning Fallacy), practice egoless programming, know the importance of (sometimes irrational) feelings people have.
- Polymorphism in Clojure – Tim Ewald’s 1h live coding talk at Øredev conference introducing mechanisms for polymorphism (and Java interoperability) in Clojure and explaining well the different use cases for them. Included: why records, protocols & polymorphism with them (shapes, area => open, not explicit switch) (also good for Java interop.: interfaces), reify, multimethods.
- Stuart Sierra: Thinking in Data (1h talk) – Sierra introduces data-oriented programming, i.e. programming with generic, immutable data structures (such as maps), pure functions, and isolated side-effects. Some other points: Records are an optimization, only for perforamnce (rarely) or polymorphism (ot often); the case for composable functions; testing using simulations (generative testing) etc.; visualization of state & process
Tools & Libs
- Netflix’ Hysterix: library to make distributed systems more resilitent by preventing a single slow/failing dependency from causing resource (thread etc.) exhaustion etc. by wrapping external calls in a separate thread with clear timeouts and support for fallbacks, with good monitoring etc. Read “Problem Definition” on the page to understand the problem it tries to solve.
if you build something that is fundamentally broken it isn’t really interesting that you followed the plan or you followed some methodology — the thing you built is fundamentally broken.
– Bjørn Borud, Chief Architect at Comoyo.no, in an email 12/2012
The root of the Toyota Way is to be dissatisfied with the status quo; you have to ask constantly, “Why are we doing this?”
– Katsuaki Watanabe, Tyota President 2005 – 2009 (from the talk Deliberate Practice)
Posted in General, Languages, Tools, Top links of month | Tagged: aws, bigdata, clojure, cloud, development, java, kent beck, methodology, outsourcing, scala, scaling, webapp, xp | Comments Off
Posted by Jakub Holý on October 9, 2012
To load all fields from a tab-separated text file in Cascalog we need to use the generic hfs-tap and specify the “scheme” (notice that loading all fields and expecting tab as the separator is the default behavior of TextDelimited):
With a custom separator and fields:
(cascading.scheme.hadoop.TextDelimited. (cascalog.workflow/fields ["?f1" "?f2"]) "\t") ; or cascading.tuple.Fields/ALL inst. of (fields ...)
Hadoop doesn’t manage to load data files from nested sub-directories (for example from a Hive partitioned table). To load them, you need to use a “glob pattern” to turn the standard Hfs tap into a GlobHfs tap. This is how we would match all the subdirectories (Hadoop will then handle loading the files in them):
Posted in General, Languages, Tools | Tagged: bigdata, cascalog, clojure, java | Comments Off
Posted by Jakub Holý on September 21, 2012
Hadoop’s NameNode and JobTracker expose interesting metrics and statistics over the JMX. Hive seems not to expose anything intersting but it still might be useful to monitor its JVM or do simpler profiling/sampling on it. Let’s see how to enable JMX and how to access it securely, over SSH.
Read the rest of this entry »
Posted in Tools | Tagged: bigdata, hadoop, hive, monitoring, ops | 2 Comments »
Posted by Jakub Holý on August 8, 2012
I was surprised not to be able to google an answer to this so I want to record my findings here. To add (a.k.a. commision) a node to Hadoop cluster that should be used only for map-reduce tasks and not for storing data, you have multiple options:
- Do not start the datanode service on the node
- If you’ve configured Hadoop to allow only nodes on its whitelist files to connect to it then add it to the file pointed to by the property mapred.hosts but not to the file in dfs.hosts.
- Otherwise add the node to the DFS’ blacklist, i.e. file pointed to by the property dfs.hosts.exclude and execute
hadoop dfsadmin -refreshNodes on the namenode to apply it.
Read the rest of this entry »
Posted in General, Tools | Tagged: bigdata, hadoop, ops | Comments Off